CISA Acting Director uploads sensitive documents to public ChatGPT instance
CISA Acting Director Madhu Gottumukkala inadvertently exposed sensitive but unclassified government contracting documents by uploading them to a public ChatGPT instance. The breach was detected by internal security sensors and led to a DHS-level damage assessment.
The acting head of the nation's cyber defense agency bypassed secured federal AI tools to use a public instance, risking the exposure of sensitive government data.
Key facts
- What
- CISA Acting Director Madhu Gottumukkala inadvertently exposed sensitive but unclassified government contracting documents by uploading them to a public ChatGPT instance.
- Incident date
- Jul 15, 2025
- Who
- Cybersecurity and Infrastructure Security Agency (CISA)
- Failure mode
- Data Leakage
- AI surface
- Chatbot
- Severity
- High
What happened
Acting Director Madhu Gottumukkala uploaded sensitive government contracting documents marked for official use only into a public version of ChatGPT in mid-July 2025. The uploads triggered multiple automated security warnings within CISA's monitoring infrastructure in early August. DHS leadership subsequently conducted a damage assessment to evaluate the risk to government security.
What broke inside the model
- 01 · TriggerA request triggers retrieval or context loading.
- 02 · Model stepThe context pulls in another user's content.
- 03 · Control gapNo boundary enforces isolation at the moment of output.
- 04 · FailurePrivate data crosses into the response.
- 05 · ConsequenceOne user sees another's data, and disclosure follows.
One user's content crosses the retrieval boundary into another's response.
The failure occurred because a public AI tool was used for processing sensitive government data, which lacks the isolated environment required for for official use only material. Public ChatGPT instances store user-submitted content to improve model responses, creating a risk of data exfiltration to a commercial third party.
What it cost
Sources
- PressTrump’s acting cyber chief uploaded sensitive files into a public version of ChatGPTpolitico.com
- PressCISA Chief Accidentally Uploads Sensitive Government Documents to Public ChatGPTcyberpress.org
- PrimaryGrassley to CISA - ChatGPTgrassley.senate.gov
Cite this entry
https://failureindex.ai/failures/cisa-acting-director-uploads-sensitive-documentsAI Failure Index. "CISA Acting Director uploads sensitive documents to public ChatGPT instance" (FI-0620). Realm Labs. https://failureindex.ai/failures/cisa-acting-director-uploads-sensitive-documents (indexed Jun 22, 2026).Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0620. Full dataset at /data.
Note from Realm Labs, the Index steward
How Realm would have caught this
- Prism
- OmniGuard
- AI Detection & Response (AIDR)
Realm can detect when a response is about to emit data that falls outside the bounds of the current user and context, and block or redact it inline, at the moment of generation rather than after the data has left.